Development of explainable prediction model for major depressive disorder based on resting-state causal connectome

Abstract

and HCP, Human Connectome Project: dataset of 865 subjects) were used to implement a depression-prediction model, based on machine learning and deep learning. A graph neural network-based model capable of learning directed causal-graph properties showed the best performance

its explanatory model suggested relevant brain regions and the connectivity of depression. The framework applied in this study is expected to shed light on the pathophysiology of depression at both the individual and group levels, providing assistance for diagnosis and therapeutic decision-making in clinical practice.

In recent decades, many attempts have been made to identify biomarkers and build prediction models for depression, using the resting-state functional connectome. However, most of these have used brain connectivity, based on temporal correlation and small-scale datasets

they are therefore vulnerable to confounding effects. The present study has obtained a whole-brain causal connectome from the brain’s resting-state functional MRI signals. Data drawn from three large-scale datasets (YAD, Young-Age Depression: dataset of 294 participants

EMBARC, Establishing Moderators and Biosignatures of Antidepressant Response in Clinical Care: dataset of 162